Randy Jensen, Stottler Henke Associates, Inc.; Arthur Goldblatt, Stottler Henke Associates, Inc.; Joseph Birtman, Stottler Henke Associates, Inc.; Devin Cline, Stottler Henke Associates, Inc.; William Lin, Stottler Henke Associates, Inc.; Gregory Howe, Stottler Henke Associates, Inc.
Keywords: Space Domain Awareness, Multi-Source Reasoning, Data Fusion, Relational Network, Knowledge Graph, Satellite Characterization, Object Classification, Similarity-Based Inference
Abstract:
Defending space systems against intentional and unintentional threats depends greatly on space domain awareness (SDA), facilitated by tools to help human operators make faster decisions. With the proliferation of space objects there has also been an increase in sensors, resources, and analytics to collect, process, and reason from observational data to support multiple levels of situational awareness, from perception to comprehension, and then projection. And yet, uncertainty remains a persistent challenge that space operators will continue to face. Human operators must often rely on informed judgments about likelihoods, based on available data and knowledge of historical behaviors and patterns. In this paper we describe an automated similarity-based inferential approach to help space operators gain better situational awareness at the comprehension level, and ultimately avoid operational surprise. Using a knowledge graph constructed as a relational network, automated reasoning about space objects is based on calculated similarities and relationships. This approach has been implemented in a functional tool initially tested in the SDA Tools Applications and Programming (TAP) Lab as part of an integrated battle management framework under development.
The inferential tool uses a relational network with nodes for space objects, launch events, launch vehicles, countries, orbital parameters, and other space-related features. Extended multi-source data include nodes for characterizations of dynamic events. Relationships between nodes are represented as edges. The network uses a similarity function that works like a “friend recommender” in social network systems, based on the strength of edges between nodes, together with feature analysis within nodes. This allows for automated inferences in ways that are comparable to how human operators reason about unknowns regarding space objects.
This approach supports machine-to-machine automation for SDA analytics, using standardized methods for data throughput, such as the Unified Data Library (UDL), and the database and messaging utilities in the SDA TAP Lab. Integration with other tools includes the ingestion of catalog data from Space-Track and open sources. Upstream tools also provide further catalog data as well as event data and indicators produced by analytic algorithms to help characterize meaningful behaviors such as a maneuver or an unusual photometric change. Similarity-based inferences are passed as outputs to downstream tools, including: (i) objects identified as similar to known threats, (ii) likely values for unknowns, such as the likely identity or constellation for unnamed objects in a catalog, or likely capabilities and behaviors, and (iii) multi-object relationships such as functional siblings or newly launched payloads working in conjunction with existing objects on orbit. Along with the implementation of the relational network, a human-machine interface supports operator investigation to visualize the network and see patterns or clusters of similarities. This provides explainability as well, by making the justification for inferences apparent in terms of which objects are similar or related, and why. Additional user interface utilities support filter, search, and customization of the feature-wise weights used in calculating similarity.
This paper describes the technical approach with examples based on selected use cases, and walks through the current applied implementation and interoperability within the SDA TAP Lab battle management architecture, as well as plans for future development.
Date of Conference: September 16-19, 2025
Track: Space Domain Awareness